longitudinal Predictors of aerobic Performance in adolescent soccer Players

João Valente-dos-santos1, Manuel J. Coelho-e-silva1, João Duarte1, antónio J. Figueiredo1, João R. liparotti2, lauren B. sherar3, Marije t. elferink-gemser4, 5, Robert M. Malina6, 7 1Faculty of Sport Sciences and Physical Education, University of Coimbra, Portugal, 2Department of Physical Education, Center of Health Sciences, Federal University of Rio Grande do Norte, Brazil, 3School of Sport, Exercise and Health Sciences, Loughborough University, Leicestershire, United Kingdom, 4Center for Human Movement Sciences, University Medical Center Groningen, University of Groningen, the Netherlands, 5Institute for Studies in Sports and Exercise, HAN University of Applied Sciences, Nijmegen, the Netherlands, 6Department of Kinesiology and Health Education, University of Texas, Austin, United States, 7Department of Health and Physical Education, Tarleton State University, Stephenville, United States


Introduction
Match analysis of elite soccer has indicated that high-intensity actions are important in soccer (1), but with players covering an average of approximately 8-12 km during a single match (2), soccer is largely dependent upon a high level of aerobic capacity (3). Increases in absolute maximal oxygen uptake in adolescents are strongly correlated with increases in body weight, which, among other factors, are connected to changes in the lungs, heart, and skeletal muscle during puberty (4). Variation in size and performance associated with interindividual differences in biological maturation are especially important during the transition into and during adolescence in male soccer players. Players advanced in skeletal age (SA) relative to chronological age (CA) (higher SA/CA ratio) tend to be taller, heavier, stronger, more powerful, and faster than players somewhat "delayed" in skeletal maturation (5).
Relationships between biological (i.e., sexual) maturity and maximal O 2 uptake in youth soccer players (12-19 years) have been previously addressed (6). Results suggested that biological maturity did not exert an effect on aerobic fitness. Previous studies were cross-sectional (3,5,(7)(8)(9) and consequently did not permit a clear distinction between the effects of growth, maturation and the effects of training on functional capacities. Longitudinal studies that simultaneously explore the effects of changes and relative contribution of body size and composition, biological maturation, and training and competition on aerobic capacity development in youth soccer players are needed.
It has been shown that the measured maximal O 2 uptake of professional soccer players do not change before and after a 5-week aerobic training period; however, time to exhaustion on the maximal O 2 uptake test on the treadmill increased significantly by 12% (10). These findings suggested that the measured maximal O 2 uptake was not a sensitive indicator of developmental changes in mid-and long-term maximal efforts in soccer players. Performance in a 20-m multistage continuous shuttle run is a motor task that also combines coordination, lower limb strength, and motivation, and it has been reported to be strongly correlated with the directly measured maximal O 2 uptake (r=0.80) suggesting that it could be used as a surrogate measure of aerobic capacity in children (11).
Longitudinal observations in soccer players suggested that maximal gains in aerobic performance occurred, on average, close to the time of peak height velocity, the interval of maximal growth rate in height during the adolescent growth spurt (12). The corresponding cross-sectional data of adolescent soccer players showed that Portuguese players 11-12 and 13-14 years of contrasting skeletal maturity status did not differ in speed, agility, power and four soccer skills (i.e., ball control, dribbling, speed, shooting accuracy, wall pass), but did differ in aerobic performance assessed by a 20-m intermittent shuttle run protocol, with "early" maturing athletes performing better than "on time" and "late" maturers (7). Our knowledge of the longitudinal relationships among aerobic performance, growth, maturation, and training is limited. Therefore, the purpose of this study was to evaluate the longitudinal development of the aerobic performance of youth soccer players aged 10 to 18 years, with a specific emphasis on the contributions of CA, SA, body size, and training as the potential explanatory variables.

Material and Methods
Subjects and Procedures. The research proposal was approved by the Scientific Committee of the University of Coimbra and the Portuguese Foundation for Science and Technology (PTDC/ DES/112781). The Portuguese Soccer Federation and clubs were contacted, and institutional agreements were signed between the University of Coimbra and the sport organizations to assure the 5-year data collection. The guardians of the young athletes provided informed consent, and players provided their assent. Goalkeepers were not included since they complete different tasks during the game compared to outfield players.
The sample included 83 players aged 11-13 years at baseline from 5 local clubs in the Midlands of Portugal who were followed on an annual basis over 3 to 5 years (mixed-longitudinal). An average of 4.4 observations per player (min, 3; max, 5) were available. Each year data were collected within a 2-week period separated by 1 year under standard conditions at an indoor facility at the University of Coimbra. Assessments, including radiographs of the left hand-wrist, were performed at the same time of day (6:00 pm to 7:00 pm) during April.
Training History. The clubs were involved in a 9-month competitive season (September-May) reg-ulated by the Portuguese Soccer Federation. Teams had 3-5 training sessions per week (90-120 minutes per session) and one game, usually on Saturday. The players also participated in competitions at the national level, and it can be described as a developmental pool for a potential talent identification. Years of formal participation in soccer for each player were obtained through the publicly available Portuguese Soccer Federation records and were verified by club records. The annual volume of training (number of sessions and minutes) was recorded for each player by a research assistant. Playing time (minutes and games) were also collected, but not considered in the current study.
Age and Skeletal Maturity. Decimal CA was calculated as the difference between date of birth and date of the hand-wrist radiograph, which was used to assess SA. The posterior-anterior radiographs of the left wrist were taken, and the films were rated using the Fels method for the assessment (13). The protocol assigns grades to specific maturity indicators for the radius, ulna, carpals, metacarpals plus phalanges of the first, third, and fifth rays and utilizes the ratios of linear measurements of the widths of the epiphysis and metaphysis of the long bones. The presence (ossification) or absence of the pisiform and adductor sesamoid bones is also noted. The grades and the ratios were entered into a program (Felshw 1.0 Software, Lifespan Health Research Center, Departments of Community Health and Pediatrics, Boonshoft School of Medicine, Wright State University, Dayton, Ohio) to derive each subject's SA. The statistical protocol weighted the contributions of specific indicators, depending on CA and sex, in calculating SA and its standard error of estimate (a confidence interval for the assessment) (14,15).
Anthropometry. A single anthropometrist measured stature, body weight, and 2 skinfolds (i.e., triceps and subscapular) following the standard procedures (16). Stature was measured to the nearest 0.1 m using a Harpenden stadiometer (model 98.603, Holtain Ltd, Crosswell, UK), and body weight was measured to the nearest 0.1 kg using SECA scales (model 770, Hanover, MD, USA). Skinfolds were measured to the nearest mm using a Lange caliper (Beta Technology, Ann Arbor, MI, USA). The technical errors of a measurement for stature (0.27 cm), body weight (0.47 kg), and skinfolds (0.47-0.72 mm) were well within the range of several health surveys in the United States and a variety of field surveys (17). Body fat was estimated from triceps and subscapular skinfold thicknesses using the protocol of Slaughter et al. (18). Fat-free mass (FFM) was derived in kg.
Aerobic Performance. Aerobic performance was measured using the 20-m multistage continuous shuttle endurance test (19), a standard field test included in the European fitness test battery (20) and in the Portuguese physical education curriculum. In brief, 5-10 athletes performed a series of runs across a 20-m track, changing direction at the end of each run to coincide with an audio signal that was getting progressively faster. Subjects started running at a speed of 8.5 km/h, and the speed increased at various stages (0.5 km/h every minute). Each stage was made up of several shuttle runs, and players were instructed to keep pace with the signals as long as possible. The results were recorded as laps taken to complete the 20-m shuttle run test. Aerobic performance was expressed as the number of completed laps achieved in the shuttle run test. The test-retest (average 7 days apart) reliability (21) in 32 players was high (R=0.86).
Statistical Analysis. Means and standard deviations were calculated for the repeated measures of CA, SA, training history, anthropometrics, and aerobic performance.
Multicolinearity was examined using a correlation matrix and diagnostic statistics. Variables with small tolerance (≤0.10) and a variance inflation factor of >10 (corresponding to an R 2 of 0.90) are generally considered indicative of harmful multicolinearity (22). The incidence of a nearly perfect bivariate correlation between body weight and FFM (r=0.96) and a very large bivariate correlation between CA and SA (r=0.83) suggested an unacceptable multicolinearity occurrence. To avoid harmful multicolinearity, body weight was discarded by the auxiliary regression, and 2 multilevel models of predictors were considered. The Model 1 adopted CA as a level 2 variance component, and the Model 2 considered SA. Additionally, Pearson product moment correlation coefficients were used to determine the relationship between the dependent variable and the possible explanatory variables, establishing the order of entrance in the multilevel analysis (CA, r=0.64; SA, r=0.53; FFM, r=0.49; stature, r=0.47; annual volume training, r=0.36). Correlations were considered as trivial (r<0.1), small (0.1<r<0.3), moderate (0.3<r<0.5), large (0.5<r<0.7), very large (0.7<r<0.9), and nearly perfect (r>0.9) (23).
For the longitudinal analyses, a hierarchical random-effects model was constructed using a multilevel modeling approach (MLwiN 2.02). Multilevel modeling effectively captures the feature that the variance of the observations increases with time, and because each individual has his or her own slope and intercept, it provides the opportunity to determine the effects on the slope and intercept of each predictor variable and its significance by relating the observed effects to the respective standard errors (24). Thus, group effects larger than withinindividual variation can be identified.
Moreover, it is common in longitudinal studies for some individuals to miss an observation and for some to drop out. The multilevel model has an advantage since the number of observations and temporal spacing between measurements can vary among subjects; all available data can thus be incorporated into the analysis. The multilevel models assume that the probability of missing data is independent of any of the random variables in the model. As long as a full information estimation procedure is used, such as maximum likelihood in ML-wiN for normally distributed data, the actual missing mechanism can be ignored (25).
The repeated measurements of aerobic performance development were assessed within an individual (Level 1 of the hierarchy) and between individuals (Level 2 of the hierarchy). A detailed description of multilevel modeling and complete details of this approach are presented elsewhere (26). An additive polynomial random-effect multilevel regression model (25) was adopted to describe the developmental changes in aerobic performance. In a first attempt, the constant and CA (Model 1, CA as Level 2 predictor) and the constant and SA (Model 2, SA as Level 2 predictor) were allowed to vary randomly between individuals (Level 2). However, CA or SA as our time-dependent variables dramatically increased the parameter estimate of variance at Level 2, around the between-individuals intercept. This is because all individuals have different developmental performance trajectories. To overcome this problem, it was decided to shift the origin of the explanatory random variables (CA and SA) by centering to their mean values (i.e., CA 14.21 years and SA 14.56 years) (26).
Subsequently, predictor variables were accepted as significant if the estimated mean coefficient was greater than twice the standard error of the estimate (P<0.05). The final model included only variables that were significant independent predictors. To allow the nonlinearity of the aerobic performance development, age power functions (i.e., centered CA 2 and centered SA 2 ) were introduced into the linear model (24). The alpha level was set at 0.05.

Results
During the 5-year testing period, 366 observations on the outcome variable and all potential predictors were performed ( Table 1). As expected, the mean values increased with age. The mean SA always exceeded the mean CA within each age cohort, and SDs ranged from 1.0 to 1.3 years in players aged 11 and 16 years. Tables 2 and 3 summarize the results from the multilevel models. The random-effects coefficients describe the 2 levels of variance (within individuals [Level 1 of the hierarchy] and between individuals [Level 2 of the hierarchy]). The significant variances at Level 1 indicated that the aerobic performance increased significantly at each measurement occasion within individuals (estimate, >2×SE; P<0.01). The between-individuals variance matrix (Level 2) indicated that individuals had significantly different aerobic performance growth curves in terms of both their intercepts (constant/constant, P<0.01) and slopes of their lines (age/age, P<0.01). The variances of these intercepts and slopes were negatively and nonsignificantly correlated (constant/ age, P>0.05) in the multilevel model. The negative sign of the covariance between intercepts and slopes means that at older age, the improvement of the aerobic performance occurs at a lower rate, and the lack of correlation suggests that individuals with higher intercepts do not necessarily have steeper slopes.   After each explanatory variable was adjusted for covariables, it can be seen that in the multilevel Model 1 (CA as Level 2 predictor, Table 2), centered CA (P<0.01), centered CA 2 (P<0.01), and annual volume training (P<0.05) had significant effects on the aerobic performance of these soccer players. The multilevel Model 2 (SA as Level 2 predictor, Table 3) included centered SA (P<0.01), centered SA 2 (P<0.05), FFM (P<0.01), and annual volume training (P<0.05). The power function of centered CA (i.e., centered CA 2 ) allowed shaping individual curves and then making the model nonlinear. It was also possible to develop equations to obtain estimates of the aerobic performance of young soccer players through the use of the multilevel models considering chronological or biological age: Model 1=57.75+9.06×centered CA-0.57×centered CA 2 +0.03×annual volume training; Model 2=13.03+4.04×centered SA-0.12× centered SA 2 +0.99×FFM+0.03×annual volume train ing. For example, in both models, one additional hour of annual training predicts, approximately, a 0.6-m improvement in the aerobic performance test (0.03×20 m).
The predicted mean aerobic performance scores aligned by CA (AE CA ) and SA (AE SA ) were plotted ( Fig.). The predicted values improved with age. Nevertheless, after 15 years of age, small improvements were noted in AE CA . Despite the lower annual gains after the age of 12, the APSA improved continuously until the age of 18.

Discussion
This study developed a multilevel model that better predicts the aerobic performance of adolescent male soccer players taking into account the athlete's chronological age, biological age (i.e., SA), physique, and training. To our knowledge, this is a novel approach. Through the use of multilevel modeling, we were able to model not only the withinsubject variation of the aerobic performance with time, but also the variation between subjects while simultaneously accounting for other explanatory variables, such as CA, biological age, and training, of which have been shown to contribute to withinsubject variation. The multilevel modeling procedure enabled us to separate out the effects of biological maturation while avoiding the errors inherent when using ratio standards and provided the means to study to what extent each factor influenced the pattern of development of the aerobic performance.
A recent longitudinal study (27) investigated the development of the intermittent endurance capacity in a sample of 130 talented Dutch soccer players aged 14-18 years who became professional and nonprofessional in adulthood. Even though the interactions with growth and maturation were not considered, players who reached the professional league showed a differential development pattern compared to their counterparts, and similarly to our results, reported a direct relationship between intermittent endurance capacity and time spent in soccer activities (training plus independent practice). It is known that with growth and training, players can improve their performance by increasing aerobic output during a particular movement (28). Young adults typically show a 15%-20% increase in maximal O 2 uptake with training, although there may be a large intraindividual variation due to genetic factors (29). Despite the arguments and research suggesting that prepubescent children are not capable of improving their endurance performance with training (9), there is much evidence to suggest otherwise (30).
Although the effects of exercise training on prepubescent children remain debatable (31), the prospective part of this study revealed that 1 extra hour of soccer-specific training represented an 0.6-m improvement in the aerobic performance of our athletes. This means that if these players have a regular training season of about 36 weeks (~9-month competitive season) and spend 1 hour extra per week training, a 22-m improvement would be expected. Although Williams and Reilly (32) noted that physiological measurements could not be used reliably on their own for the purposes of talent identification and selection, they also argued that physiological characteristics (such as aerobic performance) might be more influential in successful performance in the future, since contemporary professional soccer will be played at a higher tempo.
Longitudinal data on boys from the Leuven Longitudinal Twin Study suggested a simultaneous regulation of the timing of maximum growth in body dimensions and aerobic fitness during adolescence (33). The results from the Training of Young Athletes (TOYA) study indicated that the maximal O 2 consumption, adjusted for age and body dimensions, increased with a pubertal status in male ath-  (24). In general, the available data indicate that age-related increases in the maximal O 2 consumption are mostly mediated by changes in size dimensions, as the hematological components of oxygen delivery and the oxidative mechanisms of the exercise muscle are related to body dimensions and muscle mass (34,35). However, the individuality of timing and the tempo of maturation and year-to-year changes in body weight and the maximal O 2 consumption may be masked by maturity effects (17).
Several studies concluded that physical performance is related to anthropometric characteristics, partly because of growth and maturation (5,8,35). However, anthropometric characteristics are also related to CA, and FFM and the percentage of body fat can be influenced by training (17). A recent investigation using multilevel modeling showed that within the age range of 11-18 years, both CA and the stage of maturity were explanatory variables of maximal O 2 uptake, independent of body size and fatness (36). In the present study independent of the influence of the anthropometric characteristics, CA and SA had a significant contribution to the improvement of the aerobic performance. Additionally, the importance of variability associated with the biological age on the aerobic performance should be highlighted, even in the latter period of adolescence.
Adolescent male athletes tend to be advanced in skeletal maturation (5,7,17). In support of this, the boys in the present sample were advanced in SA relative to CA (Table 1). Changes in aerobic performance when aligned by CA and SA showed 2 distinct patterns (Fig.). The model aligned by CA suggested an initial linear development in aerobic performance toward the end of puberty. After the age of 15 years, the relationship with CA develops in a curvilinear fashion with a lower rate of improvement. This is consistent with the average age of end of peak growth velocity in aerobic endurance (12). The considerable interindividual variation in the SA of athletes of the same CA suggests a more complex relation between biological age and aerobic performance. Specifically, the development of the aerobic performance proceeds nearly linearly between 10 and 18 years of age. Whether this development proceeds linearly after the SA of 18 years is unknown. Future research could examine the development of the aerobic performance of soccer players after age 18 to clarify this issue.
From a practical point of view, routine testing of aerobic performance is easier when it is based on CA. Unavoidably, players who are older have higher endurance than younger individuals; however, the differences in athletic performance should be analyzed carefully because individuals of the same age may show large variations in biological age. This can largely confound the process of talent identification during the adolescent years. Although this study reports an important biological explanation of the aerobic performance, SA assessments are costly and require specialized equipment and interpretation, which may hinder its use in some field situations. Despite the limitation of noninvasive maturity estimates, methods like the prediction of percentage of adult height attained (without SA in the equation) (37) and the prediction of age at peak height velocity using anthropometric measurements (38) may be potentially useful tools in the practical context of sport. It is uncertain if our results can be generalized to other soccer populations, but it should be noted that there was only a slight selection bias as we had a good response rate.

Conclusions
The present study provided relevant developmental models to interpret intra-and interindividual variability in aerobic performance among adolescent soccer players. Aerobic performance was substantially related to chronological age, its power function (i.e., CA 2 ), and annual volume of training. In parallel, when considering skeletal age, the influence of fat-free mass and annual volume of training on aerobic performance was also apparent. In summary, our models can be used by coaches and physical conditioners to interpret the magnitude of annual gains in aerobic performance of adolescent soccer players.